Dokumen ini tidak dikendalikan jika diunduh Uncontrolled when downloaded 2.4
Approaches to the Estimation of Uncertainty
There are various published approaches to the estimation of uncertainty andor variability in testing. ISOIEC 17025 does not specify any particular approach.
Laboratories are encouraged to use statistically valid approaches. All approaches that give a reasonable estimate and are considered valid within the
relevant technical discipline are equally acceptable and no one approach is favoured over the others. The following are examples of approaches.
a. Both the intermediate precision and reproducibility from inter- laboratory
comparisons described in ISO 5725
5
see clause 5.4.6.3, note 3 of ISOIEC 17025 may be used in estimating measurement uncertainty.
However, these may omit some uncertainty sources that should also be estimated and combined with the precision, if significant.
b. Guide to the Expression of Uncertainty in Measurement GUM
6
see clause 5.4.6.3, note 3 of ISOIEC 17025 is often regarded as having the
more rigorous approach to the estimation of uncertainty. However, in certain cases, the validity of results from a particular mathematical
model may need to be verified, e.g. through inter-laboratory comparisons.
c. In those cases where a well-recognised test method specifies the limits
to the values of the major sources of uncertainty of measurement, and specifies the form of presentation of calculated results, the laboratory
can be considered to have satisfied the uncertainty of measurement requirements see clause of 5.4.6.2, note 2 ISOIEC 17025 by following
that test method.
2.5 Degree of Rigour
The degree of rigour and the method used for estimating uncertainty should be determined by the laboratory in accordance with note 1 of clause 5.4.6.2 of
ISOIEC 17025.
To do this, the laboratory should: a.
consider the requirements and limitations of the test method and the need to comply with
“good practice” in the particular testing sector; b.
ensure that it understands the requirements of the customer see clause 4.4.1 a of ISOIEC 17025. It is often the case that the
customer understands the problem but does not know what tests are required, and thus needs guidance on the uncertainty required for
solving the problem;
c. use methods, including methods for estimating uncertainty, which meet
the needs of the customer see clause 5.4.2 of ISOIEC 17025. It should be noted that what a customer wants may not be what is
appropriate for the testing under consideration;
Dokumen ini tidak dikendalikan jika diunduh Uncontrolled when downloaded
d. consider the narrowness of limits on which decisions on conformance
with specification are to be made; e.
consider the cost effectiveness of the approach adopted. In general, the degree of rigour relates to the level of risk. To properly
evaluate safety or substantial property risk or financial risk, a relatively rigorous uncertainty estimate is required for the associated tests or
measurements. For property evaluations where the test result supports a
“fitness for purpose” conclusion, the associated test or measurement uncertainty may have a minor effect on the conclusion and would thus require
a less rigorous estimate.
In general, if less rigour is exercised in estimating measurement uncertainty, the estimated measurement uncertainty value will be larger than an estimate
obtained from a more rigorous approach. Semi-quantitative measurements require less rigorous treatment of measurement uncertainty.
If there is a large margin between the measured results and the specified limits, test results with a larger uncertainty are acceptable and a less rigorous
approach to the estimation of measurement uncertainty can be justified. See 1.1.6 of APLAC TC004
7
for a more detailed discussion on how measurement uncertainty affects the ability to distinguish compliance from non-compliance.
If the estimated uncertainty is not acceptable to the laboratory ’s customer or
is too large for determination of compliance with the specification, the laboratory should endeavour to reduce the uncertainty, e.g. through
identification of the largest contributors to uncertainty and working on reducing these.
2.6
Uncertainty Arising from Sampling
Measurement uncertainty strictly applies only to the result of a specific measurement on an individual specimen.
During contract review there should be consideration and agreement with the customer as to whether the test result and uncertainty are to be applied to the
specific sample tested or to the bulk from which it came. Where sampling or sub-sampling is to be treated as part of the test, the
uncertainty arising from such sampling should be considered by the laboratory. Estimating the representativeness of a sample or set of samples
from a larger population requires understanding of the homogeneity of the larger population and additional statistical analysis.
Where a test method includes specific sampling procedures designed to characterise a batch, lot or larger population, the measurement uncertainties
for individual measurements are often insignificant relative to the statistical variation of the batch, lot or larger population. In cases where the
measurement uncertainty of individual measurements is significant in relation to the standard deviation of the sampling, the measurement uncertainty
should be taken into consideration when characterising the batch, lot or larger population.
Dokumen ini tidak dikendalikan jika diunduh Uncontrolled when downloaded
Where the test procedure includes a specific sub-sampling procedure, it is necessary to analyse the representativeness of the sub-sample as part of the
measurement uncertainty estimation. Where there is doubt about the representativeness of a sub-sample, it is recommended that multiple sub-
samples be taken and tested to evaluate the homogeneity of the prepared sample from which sub-samples were drawn.
Where only one sample is available and is destroyed during the test, the precision of sampling cannot be determined directly. However, the precision
of the measurement system should be considered. A possible method for estimation of the precision of sampling is to test a batch of
“homogeneous” samples for a highly repeatable measurand and to calculate the standard
deviation of sampling from the results obtained.
2.7 Reporting Measurement Uncertainty
